Statistics hub · 91 data points · 26 primary sources
AI Coding Tools Statistics 2026: 90+ Verified Numbers on Adoption, Productivity, Quality and Benchmarks
We collected and verified these numbers so you don't have to. 90+ statistics on AI coding assistants and coding agents from primary sources wherever they exist (secondary reports are marked) (Stack Overflow, GitHub Octoverse, DORA, METR, Faros, Bain, Atlassian, Veracode, GitClear, SWE-bench, Google Research, Anthropic) — each linked to its report, sample size and date, so any team can base build, buy and hiring decisions on them.
Why this page exists: the AI coding debate runs on anecdotes and vendor benchmarks. We read the original surveys, telemetry studies and randomised trials, kept only numbers traceable to the organisation that measured them, and put the contradictory ones side by side. The pattern that emerges — adoption near-universal, self-reported gains large, measured gains small or negative, quality costs real — is more useful than any single headline.
Adoption
Usage is near-universal in surveys and telemetry alike. The differentiator is now depth — daily agent use vs occasional autocomplete — and whether time saved goes anywhere.
- 84% of developers use or plan to use AI tools in their development process (76% in 2024, 70% in 2023) [1]
- 47.1% / 17.7% / 13.7% use AI tools daily / weekly / monthly or less; 16.2% do not plan to [1]
- 50.6% of professional developers use AI tools daily [1]
- 14.1% of developers use AI *agents* daily; 37.9% have no plans to [1]
- 82% of AI-tool users use them to write code; 67.5% to search for answers; 56.7% for debugging; 27.2% for testing [2]
- 90% of ~5,000 technology professionals report using AI at work [3]
- ~80% of new developers on GitHub use Copilot within their first week [4]
- 36M+ new developers joined GitHub in the year to Aug 2025 (+23% YoY), a step-change coinciding with Copilot Free [4]
- 50% of open-source projects have at least one maintainer using GitHub Copilot [4]
- 1M+ pull requests created by the Copilot coding agent between May and Sep 2025 [4]
- 1.13M+ public repositories import an LLM SDK (+178% YoY); 693,867 were created in the past 12 months [4]
- >60% weekly-active-user threshold for 'widespread adoption' — reached by most companies only in the last 2–3 quarters before June 2025 [7]
- 2 in 3 software firms have rolled out generative AI tools; developer adoption within them is low [8]
- 25% of the YC W25 startup batch had codebases that were ~95% AI-generated [15]
- 72% of developers say vibe coding is not part of their professional work; a further 5% are emphatic it isn't [1]
- 81.7% / 67.9% / 40.8% of developers using out-of-the-box agents use ChatGPT / GitHub Copilot / Claude Code [1]
- 26% more completed tasks for developers given Copilot in three randomised field experiments (Microsoft, Accenture, a Fortune 100 firm; 4,867 developers); less-experienced developers adopted more and gained more [26]
Productivity: perceived vs measured
Self-reports say 10 hours a week saved. Controlled and telemetry studies say between +55% on a toy task and −19% on real work. Both are true; they measure different things.
- +19% longer to complete real issues when 16 experienced open-source developers were allowed AI tools (Cursor Pro, Claude 3.5/3.7 Sonnet), across 246 randomised tasks [5]
- −24% / −20% developers' forecast speed-up before the study / their belief after it — despite being measurably slower [5]
- −39% / −38% economics / ML experts' predicted speed-up for the same study [5]
- ~2 h average task length in the METR trial; repositories averaged 1M+ lines and 22k+ GitHub stars [5]
- 55.8% faster task completion with Copilot in GitHub's 2023 controlled experiment (implementing an HTTP server in JavaScript) [14]
- 21% / 98% more tasks completed / more PRs merged by developers on high-AI-adoption teams (10,000+ developers, 1,255 teams) [7]
- n.s. no statistically significant correlation between AI adoption and company-level throughput, DORA metrics or quality KPIs in the same dataset [7]
- 10–15% productivity boost from AI coding assistants; time saved often not redirected to higher-value work, so ROI stays flat [8]
- 25–30% productivity boost when generative AI is paired with end-to-end process redesign, vs ~10% from assistants alone [8]
- 25–35% of time from idea to launch is spent writing and testing code — the ceiling for code-generation-only gains [8]
- >80% of DORA respondents believe AI has increased their productivity [3]
- 99% / 68% of developers report time savings from AI / say they save more than 10 hours a week [9]
- 50% of developers also lose 10+ hours a week to non-coding organisational friction [9]
- 16% of developer time is spent coding [9]
- n.s. no statistically significant change in cycle time or PR throughput for ~800 developers given Copilot access [12]
- 37% acceptance rate of AI code-completion suggestions among Google engineers; 50% of code characters are AI-completed [13]
- >8% of code-review comments at Google resolved with AI assistance [13]
- 43.2M / 82.19M monthly PRs merged (+23% YoY) / pushes on GitHub in 2025, with records coinciding with Copilot agent rollouts [4]
- 26.08% increase in completed tasks (pull requests) with Copilot, pooled across three field experiments at Microsoft, Accenture and a Fortune 100 company [26]
Code quality and security
The cost side of the ledger: more duplication, less refactoring, bigger PRs, more bugs, and a security flaw rate that newer models have not improved.
| Cloned lines, share of changes | 8.3% | → | 12.3% | +4 pts (+48%) |
- 45% of code generated by 100+ LLMs across Java, JavaScript, Python and C# introduced a security flaw; newer and larger models did not improve [10]
- 41% higher bug rate among ~800 developers with Copilot access, with no change in PR throughput [12]
- 9% / 154% more bugs per developer / larger average PR size associated with AI adoption in Faros telemetry [7]
- 91% increase in PR review time on high-AI-adoption teams [7]
- 4× growth in the number of duplicated code blocks (5+ lines), 2020–2024 — a count metric, distinct from the share-of-lines figure below [11]
- 8.3%→12.3% share of changed lines that are copy/pasted, 2020→2024; 'moved' (refactored) lines fell from ~25% in 2021 to under 10% in 2024 [11]
- 151k+ repositories with Broken Access Control alerts (+172% YoY), partly from AI-generated scaffolds skipping auth checks [4]
- 37→26 days average fix time for critical vulnerabilities, 2024→2025, attributed to automation and AI tools including Copilot Autofix [4]
- 66% of developers cite 'AI solutions that are almost right, but not quite' as their top frustration; 45% say debugging AI code takes longer [1]
- 46% vs 33% distrust vs trust AI output accuracy; only 3.1% highly trust (43% trusted in 2024) [1]
- 30% of DORA respondents report little or no trust in AI-generated code — slightly lower than 2024 [3]
- 45% of professional developers say AI tools are bad or very bad at complex tasks [2]
- negative correlation between AI adoption and delivery stability in DORA 2025 — positive with throughput for the first time [3]
- 65% reduction in shortcut/loophole behaviour on agentic tasks, Claude 4 vs Sonnet 3.7 (vendor-reported) [21]
Benchmarks
SWE-bench Verified is saturated; SWE-bench Pro and Terminal-bench are the frontier. Note the private-repo drop: models do worse on code they have never seen.
- 12.47% of 2,294 SWE-bench issues resolved by SWE-agent at launch (Mar 2024) [18]
- 65% of SWE-bench Verified resolved by mini-SWE-agent, an agent implemented in 100 lines of Python (Jul 2025) [18]
- 72.5% / 72.7% Claude Opus 4 / Sonnet 4 on SWE-bench Verified (full 500 problems, no extended thinking); 79.4% / 80.2% with parallel test-time compute [21]
- 60%→~100% approximate range of the top SWE-bench Verified score during 2025, per Stanford's AI Index — the benchmark is close to saturated [22]
- 67.3 pts one-year gain on SWE-bench 2023→2024 — the largest of any tracked benchmark [23]
- 1,865 tasks in SWE-bench Pro across 41 repositories; reference solutions average 107 lines changed in 4.1 files [19]
- ~23% top-model resolve rate on SWE-bench Pro public at release (Sep 2025), vs 70%+ on Verified — GPT-5 23.3%, Claude Opus 4.1 22.7% [19]
- 23.3%→14.9% GPT-5 resolve rate, public → private (unseen commercial) repositories; Claude Opus 4.1 22.7%→17.8% [19]
- 61.5% ± 3.1 current top SWE-bench Pro public score (Muse Spark 1.1); GPT-5.4 59.1%, Claude Opus 4.6 51.9% [19]
- <10% resolve rate on the consistently hard SWE-bench Pro repositories, for all models [19]
- 43.2% Claude Opus 4 on Terminal-bench at launch (May 2025) [21]
- 88% GPT-5 (high) on Aider's polyglot benchmark, at $29 per run; DeepSeek-V3.2 reaches 74.2% for $1.30 [20]
- <3 months doubling time of agent task horizon measured on SWE-bench Verified alone, vs ~7 months across all software tasks [6]
- 7 h Claude Opus 4 ran autonomously on a Rakuten open-source refactor (vendor-reported customer claim) [21]
From autocomplete to agents
The shift in 2025–26 is from suggestions to delegation. Usage data shows a majority of coding-agent sessions are now automation, not assistance.
- 79% vs 49% of Claude Code vs Claude.ai coding conversations are automation (AI performs the task) rather than augmentation [16]
- 43.8% of Claude Code sessions are 'directive' — user hands over the whole task with minimal interaction [16]
- 27%→39% share of all Claude.ai conversations that are directive, late 2024 to mid 2025 [17]
- 31% / 28% / 14% of coding queries use JavaScript+TypeScript / HTML+CSS / Python [16]
- 32% of organisations decided against buying at least one software product because they could build it with agentic coding tools (nearly 50% among AI high performers) [24]
- 83.5% of developer agent use is in software engineering; 24.9% data/analytics; 17.6% business-process automation [1]
- 34.4% of developers building agents use MCP servers [1]
- 52% of developers say agents or AI tools changed how they work in the past year; 16.3% 'to a great extent' [1]
- 76% / 69% of developers do not plan to use AI for deployment & monitoring / project planning [1]
- 72.6% of developers using Copilot code review say it improved their effectiveness [4]
- $2B Cursor annual recurring revenue, doubled in three months (Mar 2026) [25]
- $15 / $75 Claude Opus 4 price per million input / output tokens at launch; Sonnet 4 $3 / $15 [21]
Who benefits, and the workforce
Gains concentrate in juniors and in tasks that were already rote. Senior developers on familiar codebases gain least — or lose.
- junior > senior less-experienced developers adopted Copilot more and showed larger productivity gains than senior developers in three field experiments [26]
- 5 years average prior experience of METR trial developers on their own projects — the population that slowed down [5]
- 28% vs 17% reduction in 'always-on' burnout-risk metric for developers without vs with Copilot [12]
- 3 of 4 companies say the hardest part of generative-AI adoption is getting people to change how they work [8]
- 40% maximum share of coders' work AI assistants may be able to take on (SaaS teams of 2,000–20,000 FTEs) [8]
- 63% of developers say leaders do not understand their pain points (44% said leaders were aware in 2024) [9]
- 61% vs 53% favourable sentiment toward AI tools among professionals vs those learning to code (down from 72% overall in 2024) [1]
- 75.3% of developers go to a person when they don't trust an AI answer; 61.7% for ethical or security concerns [1]
- 39% vs 14% expect AI to reduce headcount next year vs those who saw an actual AI-driven decline last year (all functions) [24]
Formulas and derived metrics
Derived quantities computed from the primary figures above. Use them to reason about agents rather than just quote them.
Perception gap
Δ_percep = t̂ − t = (−20%) − (+19%) = −39 pts Self-estimated change in completion time (t̂) minus measured change (t). In the METR trial developers believed they were 20% faster while being 19% slower — a 39-point gap. Self-reported productivity numbers (99% 'save time', >80% 'more productive') should be read with this gap in mind. [5]
Net time recovered (illustrative)
T_net = T_saved − T_lost Atlassian's 2025 survey finds 68% of developers report saving 10+ hours a week with AI, while 50% report losing 10+ hours a week to organisational friction. The two questions are answered by overlapping but not identical respondents, so this is not a precise subtraction — but for a typical respondent the two are of the same order of magnitude. Bain reaches a similar conclusion: 10–15% gains from assistants alone rise to 25–30% only when the surrounding process changes. [9]
Review load multiplier
R = (1 + ΔPRs) × (1 + Δsize) = 1.98 × 2.54 ≈ 5.0× Faros telemetry: high-AI teams merge 98% more PRs and PRs are 154% larger, so the volume of code entering review grows roughly 5× if both effects apply to the same teams. Review time in the same dataset rose 91%. Review, not generation, becomes the bottleneck. [7]
Expected flawed samples
E[flawed] = n × p = n × 0.45 Veracode found 45% of AI-generated code samples across 100+ models contained at least one security flaw, and larger or newer models did not improve. For n generated units, expect about 0.45·n to contain a flaw — i.e. every second AI-generated module needs a security review. [10]
Public → private repository gap
g = 1 − s_private / s_public = 1 − 14.9 / 23.3 ≈ 36% SWE-bench Pro reports resolve rates on public (open-source) and private (unpublished commercial) repositories. GPT-5 drops 36% and Claude Opus 4.1 drops 22% on private code. Task and repository differences contribute alongside training-data exposure, but the direction is consistent: expect lower scores on code the model has not seen. [19]
Frequently asked questions
What percentage of developers use AI coding tools in 2026?
Stack Overflow's 2025 survey of 33,662 developers found 84% use or plan to use AI tools (76% in 2024), with 47.1% using them daily. DORA's 2025 survey of ~5,000 technology professionals found 90% use AI at work. GitHub reports ~80% of new developers use Copilot in their first week.
Do AI coding tools actually make developers faster?
It depends on who and what. GitHub's 2023 controlled experiment found 55.8% faster completion on a self-contained task; three field experiments across 4,867 developers found 26% more completed tasks. METR's 2025 randomised trial found experienced open-source developers were 19% slower on real issues in their own large repositories, while believing they were 20% faster. Bain estimates 10–15% gains from assistants alone, 25–30% with process redesign. Faros telemetry shows more PRs merged but no statistically significant company-level throughput change.
How secure is AI-generated code?
Veracode's 2025 analysis of 100+ LLMs found 45% of generated code samples contained a security flaw, and newer or larger models did not improve. GitHub's Octoverse 2025 reports Broken Access Control alerts up 172%, partly from AI scaffolds skipping auth checks.
What is the best AI coding model on benchmarks?
As of September 2026, Muse Spark 1.1 leads SWE-bench Pro public at 61.5%, followed by GPT-5.4 (59.1%) and Claude Opus 4.6 (51.9%). SWE-bench Verified is effectively saturated near 100%. On Aider's polyglot benchmark GPT-5 (high) scores 88%.
Is vibe coding common among professional developers?
Not among most professionals. In Stack Overflow's 2025 survey 72% of developers said vibe coding is not part of their professional work and a further 5% emphatically rejected it. Y Combinator reported that 25% of its W25 startup batch had codebases ~95% AI-generated.
Methodology
Primary sources are used wherever they exist: the organisation that ran the survey, benchmark or measurement. Secondary aggregators, vendor landing pages without a named methodology, and market-size forecasts without a disclosed model are excluded; the few secondary reports kept are labelled as such in the source list. Each figure is reproduced as stated by the source, with sample size and fielding dates where the source discloses them. Survey figures are self-reported by respondents and are not independently verified; benchmark figures depend on the harness and prompt used and should be compared only within a benchmark. The page is re-checked and updated as new editions of each report are published.
Sources
| # | Report | Organisation | Date | Sample / method |
|---|---|---|---|---|
| 1 | 2025 Developer Survey — AI section | Stack Overflow | 2025-07 | 33,662 developers; agent questions n=31,877 |
| 2 | 2024 Developer Survey — AI section | Stack Overflow | 2024-07 | 60,907 respondents on usage; 37,302 on trust |
| 3 | 2025 DORA Report: State of AI-assisted Software Development | Google Cloud DORA | 2025-09-24 | ~5,000 technology professionals; 100+ hours qualitative |
| 4 | Octoverse 2025 | GitHub | 2025-10-28 | Platform telemetry, Sep 2024 – Aug 2025 |
| 5 | Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity | METR (arXiv:2507.09089) | 2025-07-10 | RCT: 16 developers, 246 issues, Feb–Jun 2025; $150/h compensation |
| 6 | Measuring AI Ability to Complete Long Tasks | METR | 2025-03-19 | Models 2019–2025 on 170 software tasks |
| 7 | The AI Productivity Paradox Report | Faros AI | 2025-06 | Telemetry: 10,000+ developers, 1,255 teams; Spearman correlations, p<0.05, ≥6 companies |
| 8 | From Pilots to Payoff: Generative AI in Software Development (Technology Report 2025) | Bain & Company | 2025-09-23 | Bain client surveys and analysis; SaaS teams 2,000–20,000 FTEs |
| 9 | State of Developer Experience Report 2025 | Atlassian | 2025-07-09 | 3,500 developers and managers, six countries |
| 10 | 2025 GenAI Code Security Report | Veracode | 2025-10 | Code from 100+ LLMs in Java, JavaScript, Python, C# |
| 11 | AI Copilot Code Quality: 2025 Data Suggests 4x Growth in Code Clones | GitClear | 2025-02 | 211M changed lines, Jan 2020 – Dec 2024, enterprise and big-tech repos |
| 12 | Can Generative AI Improve Developer Productivity? | Uplevel Data Labs | 2024-09 | ~800 developers, Copilot vs control, cycle time / throughput / bug rate; cites Microsoft/Accenture field trials |
| 13 | AI in software engineering at Google: Progress and the path ahead | Google Research | 2024-06-06 | Internal engineering telemetry |
| 14 | The Impact of AI on Developer Productivity: Evidence from GitHub Copilot | GitHub / Microsoft (arXiv:2302.06590) | 2023-02-13 | Controlled experiment, recruited developers, JavaScript HTTP-server task |
| 15 | A quarter of startups in YC's current cohort have codebases that are almost entirely AI-generated (secondary report of YC's Jared Friedman) | TechCrunch, reporting Y Combinator (Jared Friedman) | 2025-03-06 | YC W25 batch, founder self-report |
| 16 | Anthropic Economic Index: AI's impact on software development | Anthropic | 2025-04-28 | 500,000 coding interactions on Claude.ai and Claude Code, 6–13 Apr 2025 |
| 17 | Economic Index report: Uneven geographic and enterprise AI adoption | Anthropic | 2025-09-15 | Claude.ai and API usage |
| 18 | SWE-bench | Princeton NLP | 2025-07 | 2,294 GitHub issues from 12 Python repos; Verified subset 500 |
| 19 | SWE-Bench Pro (Public) Leaderboard | Scale AI | 2026-09 | 1,865 tasks, 41 repos; public 731, private 276, held-out 858; ±CI reported |
| 20 | Aider polyglot coding leaderboard | Aider (Paul Gauthier) | 2025-11-20 | 225 Exercism exercises in C++, Go, Java, JavaScript, Python, Rust |
| 21 | Introducing Claude 4 | Anthropic | 2025-05-22 | Vendor-reported benchmarks; SWE-bench Verified full 500 problems |
| 22 | AI Index Report 2026 | Stanford HAI | 2026-04 | Annual compilation |
| 23 | AI Index Report 2025 | Stanford HAI | 2025-04 | Annual compilation |
| 24 | The state of AI in 2026: On the road to ROI | McKinsey & Company | 2026-08-25 | 1,719 participants, 97 countries, fielded 4 May – 8 Jun 2026 |
| 25 | Cursor Recurring Revenue Doubles in Three Months to $2 Billion (secondary; company-reported figure) | Bloomberg | 2026-03-02 | Company-reported |
| 26 | The Effects of Generative AI on High Skilled Work: Evidence from Three Field Experiments with Software Developers | Cui, Demirer, Jaffe, Musolff, Peng, Salz (Microsoft / MIT / Princeton / Wharton), SSRN 4945566 | 2024-09 | RCTs at Microsoft, Accenture and a Fortune 100 company; 4,867 developers |
Cite this page
TODO for AI. AI Coding Tools Statistics 2026. Updated September 13, 2026. https://todofor.ai/blog/stats/ai-coding-tools
@misc{todoforai_ai_coding_tools_2026,
title = {AI Coding Tools Statistics 2026},
author = {TODO for AI},
year = {2026},
url = {https://todofor.ai/blog/stats/ai-coding-tools},
note = {Updated 2026-09-13}
}Charts and figures may be reproduced with a link to this page. Individual statistics should also credit the original source listed above.